收起目录

本页内容

driver 等待请求时 kill sequence：核对后续仲裁

其他行为变化

影响范围

堆叠 sequencer，或在 get_next_item/try_next_item 等待时 kill sequence。

具体差异

旧行为——赢得仲裁的 sequence 若在 send_request 前被 kill，driver 的 get_next_item/peek 可能挂死，try_next_item 报 null item；新行为——检测 process_id.status 为 KILLED/FINISHED 时自动重新仲裁。堆叠 sequencer（sequencer 套 sequencer）需对底层 sequencer 设置 uvm_config_db#(int)::set(..., "wait_for_sequences_count", N)，否则 try_next_item 可能取不到 item。

修改方案

普通请求等待流程可保留。存在多级 sequencer 堆叠时，在目标 sequencer 的 build_phase 读取配置之前，通过 uvm_config_db#(int)::set 配置其 wait_for_sequences_count，逐级验证计数满足传播需要；kill 等待中的 sequence 后，检查下一笔请求仍能被 driver 消费。具体配置读取位置见源码依据，不把固定延迟当通用补偿。

兼容措施与范围：现有入口无需兼容开关；这不表示两版行为相同，仍按本条条件和验证方法核对。

迁移后验证

kill 等待者后再发新请求，检查重新仲裁、完成计数及等待上限；堆叠时按实际层数检查 wait_for_sequences_count。

记录本条结果：M06-012 · 状态与证据。

已有实测与复现

• Sequence kill、队列清理与 phase 生命周期：结果与复现

补充查阅

扫描定位与记录结果

有扫描规则，覆盖部分可识别写法。脚本检查名称、参数和部分 FLAG 表达式，并对少数直接声明关联使用点；未做完整类型解析或预处理，行为结果仍需验证。

• S34 · 启发式：sequence lock/kill 与等待请求流程。 排除已识别的 process/component 接收者；未知类型仍保留启发式提示。扫描不能证明并发竞争与 kill 时序。

uv run tools/uvm_migration_scan.py <SoC代码目录> --rules S34

核对命中对象及生效分支后，在条目清单记录修改与验证证据；扫描规则记录用于整理命中位置。

扫描器下载、输入范围与报告说明

源码与标准依据

核查方式：源码与实测核对。实测仅覆盖本条所列场景。

VCS 内置 UVM-1.2：uvm_sequencer.svh:230（try_next_item 一次 wait_for_sequences 后未取到 item 即报 TRY_NEXT_BLOCKED 返回 null）

VCS 内置 uvm-ieee-2020-2.0：uvm_sequencer.svh:225、uvm_sequencer_param_base.svh:474–506（新增 m_safe_select_item）、uvm_sequencer_base.svh:76/uvm_sequencer_base.svh:442–447（wait_for_sequences_count config_db 配置，默认 1）

引用代码（高亮为引用行）；上方行号链接可直接定位。

// try_next_item
// -------------

task uvm_sequencer::try_next_item(output REQ t);
  int selected_sequence;
  time arb_time;
  uvm_sequence_base seq;

  if (get_next_item_called == 1) begin
    uvm_report_error(get_full_name(), "get_next_item/try_next_item called twice without item_done or get in between", UVM_NONE);
    return;
  end

  // allow state from last transaction to settle such that sequences'
  // relevancy can be determined with up-to-date information
  wait_for_sequences();

  // choose the sequence based on relevancy
  selected_sequence = m_choose_next_request();

完整源码 · SHA256：6d2daeebf191b44e352885f8dba22a841298a245ea6facb5f7e56e1775d5ce17

// try_next_item
// -------------

task uvm_sequencer::try_next_item(output REQ t);
  int selected_sequence;
  uvm_sequence_request selected_sequence_request;
  bit found_item;
  time arb_time;

  if (get_next_item_called == 1) begin
    uvm_report_error(get_full_name(), "get_next_item/try_next_item called twice without item_done or get in between", UVM_NONE);
    return;
  end

  // allow state from last transaction to settle such that sequences'
  // relevancy can be determined with up-to-date information
  repeat (m_wait_for_sequences_count) begin
    wait_for_sequences();

完整源码 · SHA256：c4cb26775fb9b6be1a06f04045ccd8a81f7c20730e505e5e08596a47f6de096d

// m_safe_select_item
// ---------------------

task uvm_sequencer_param_base::m_safe_select_item(input bit get_next_item, output REQ t);
  process select_process;
  uvm_sequence_request selected_sequence_request;
  if(sequence_item_requested == 0) begin
    m_select_sequence(selected_sequence_request);
    fork
      begin
        select_process = process::self();
        // re-arbitrate if the sequence was killed or finished after it won arbitration but before it was
        // able to send its request.
        forever begin
          selected_sequence_request.process_id.await();

          if(!m_req_fifo.is_empty())
            break;

          if (arb_completed.exists(selected_sequence_request.request_id)) begin
            arb_completed.delete(selected_sequence_request.request_id);
          end
          m_select_sequence(selected_sequence_request);
        end
      end
    join_none
    // wait for thread to start to ensure it gets killed when peek returns
    wait(select_process != null);
  end
  sequence_item_requested = 1;
  if (get_next_item)
    get_next_item_called = 1;
  m_req_fifo.peek(t);
  if ((select_process != null) && (select_process.status != process::FINISHED))
    select_process.kill();
endtask

完整源码 · SHA256：99707a7f2e520940931db3c21cd4809bb8dfa57b324007b95501a62fbd4e1d31

  local static int              g_sequence_id = 1;
  local static int              g_sequencer_id = 1;

  protected int                 m_wait_for_sequences_count;  // specifies the # of times the sequencer
                                                             // should call wait_for_sequences().  A
                                                             // value > 1 allows sequencer stacking.
  // Function -- NODOCS -- new
  //
  // Creates and initializes an instance of this class using the normal
  // constructor arguments for uvm_component: name is the name of the
  // instance, and parent is the handle to the hierarchical parent.

  // @uvm-ieee 1800.2-2020 auto 15.3.2.1
  extern function new (string name, uvm_component parent);

  // Variable: wait_for_sequences_count
  // Controls the number of wait_for_sequences calls when selecting next sequence.
  //
  // By default, the sequencers will wait for 1 ~wait_for_sequences~ call

// -----------
function void uvm_sequencer_base::build_phase(uvm_phase phase);
  super.build_phase(phase);
  if (!uvm_config_db#(uvm_bitstream_t)::get(this, "", "wait_for_sequences_count", m_wait_for_sequences_count))
    void'(uvm_config_db#(int)::get(this, "", "wait_for_sequences_count", m_wait_for_sequences_count));

  if (m_wait_for_sequences_count < 1) begin
    `uvm_warning("UVM/SQR/WFSC", $sformatf("attempt to set wait_for_sequences_count to '%0d' will be ignored, values must be 1 or greater!", m_wait_for_sequences_count))
    m_wait_for_sequences_count = 1;
  end
endfunction : build_phase

// do_print
// --------

function void uvm_sequencer_base::do_print (uvm_printer printer);
  super.do_print(printer);
  printer.print_array_header("arbitration_queue", arb_sequence_q.size());
  foreach (arb_sequence_q[i])

完整源码 · SHA256：6d2477484529dad661a38b31619edf9898bfa3375f217f06a27b3a69af32ccf7

术语解释 · 兼容配置

返回Sequence / Sequencer · 迁移清单

版本适用范围